Training Features in Keras Policy

I’ve looked over the document but unable to understand the input, output features for Keras policy. I want to know that how the story data is featured into input features and output label for the LSTM model in Keras policy. Just show me a couple of tuples/rows for training data.

As it is based on sequential learning, so I want to know the actual sequence for training data(what it includes, only intents, or intents + utterances), and the class label for that sequence (next utterance or something else)

featurization process for policies is described here: Featurization of Conversations